The Blind First Mark

A HARVEST KERNEL WORKING SESSION

The Blind First Mark

Bring one paper from the stack you are grading this week. In about 20 minutes you will leave holding a judgment record for one criterion, written before any tool showed you a number, plus a gap rule for when a second look is required and one sentence you can tell students about how AI is used in your grading. You do the judging. Nothing here judges for you, and nothing you type leaves this page.

Why this exists. On October 13, 2026, Wonkhe published Jim Dickinson’s argument that AI marking could cost universities the protection of academic judgment. At the center of it is a randomized experiment published in PNAS Nexus: 1,339 teachers reviewed the same student work with a grade already attached, and a grade three points too harsh survived review more often when it was labeled as coming from AI. A working paper from Ofqual, the qualifications regulator in England, names the same risk, anchoring, and notes that marking in parallel avoids it. This session puts your mark down first, on one paper, before any tool gets a vote.

SEED

Pick the paper and the judgment call

Choose one submission you have not graded yet. Then choose the one criterion on your rubric where your expertise decides the mark. Not spelling, not word count, not whether the references exist. The call that only someone who knows the field can make, like the strength of an argument or the quality of an analysis.

SPROUT

Mark it blind

Before you open any grading tool, any detector or any suggested score, read the paper and decide. If a tool has already put a number on this one, cover it or pick a different paper. Your mark, the line in the student’s work that decided it, and one sentence on why. That sentence is your academic judgment, in writing, with a timestamp you can point to.

GROW

Let the tool look second

Write your gap rule first, while you still do not know what any tool will say. Then, if you use a grading tool, let it score the same criterion and compare. If the two marks sit inside your gap, your mark stands. If they do not, take a second look at the paper, not at the tool, and write down what you decided and why. If you do not use a tool, write none, and keep the gap rule for the day someone hands you one.

HARVEST

Say it out loud, then build the record

One sentence for your students about how AI is and is not used in your grading. Plain enough that a first year student understands it on the first read. If you use no tool at all, say that, because students are already wondering.

Before you copy it out, one observation.

That is one paper.

Every paper in the stack deserves the same order, your judgment first, and every one of them still needs feedback a student can actually use. Marking blind works, and it took real attention for one paper on one criterion. The judgment is the part that has to stay yours. The typing that explains it to each student is the part that does not scale by hand. It is a scale problem, not a quality problem.

When the stack is the problem

The Student Feedback Generator drafts personalized, rubric aware feedback in your tone, grounded in your Course Record: your course, your outcomes, your policies and your voice, planted once and used by every tool in the toolkit. Bring your judgment record first, then let it draft the comment. Your mark stays yours. Do the first one by hand anyway. Then you will know exactly what your feedback has to say.

See the Student Feedback Generator